StochasticDurationVitsBase<T>

Base Classes · AiDotNet.TextToSpeech.EndToEnd

A VITS-family model whose durations come from VITS’s stochastic duration predictor (Kim et al. 2021 §2.2.2), trained with the rest of the generator on the alignment’s durations and sampled through its reversed flows at synthesis.

Inheritance: NeuralNetworkBase<T> → TtsModelBase<T> → VitsTtsModelBase<T> → StochasticDurationVitsBase<T>

Known derived types: Piper<T>, VITS<T>, YourTTS<T>

For Beginners

This model learns a distribution of how long each sound lasts, so the same text can be spoken with natural variations in rhythm.

How It Works

VITS, Piper and YourTTS share this duration model; the predictor reads the stop-gradient text encoding and, when present, the stop-gradient speaker and language embeddings (reference StochasticDurationPredictor, Coqui stochastic_duration_predictor.py). Its convolutions are VitsModelOptions’ hidden width wide.